4 * Copyright (C) 1993 Linus Torvalds
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7 * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8 * Numa awareness, Christoph Lameter, SGI, June 2005
11 #include <linux/vmalloc.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/rbtree.h>
25 #include <linux/radix-tree.h>
26 #include <linux/rcupdate.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
29 #include <asm/atomic.h>
30 #include <asm/uaccess.h>
31 #include <asm/tlbflush.h>
32 #include <asm/shmparam.h>
35 /*** Page table manipulation functions ***/
37 static void vunmap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
)
41 pte
= pte_offset_kernel(pmd
, addr
);
43 pte_t ptent
= ptep_get_and_clear(&init_mm
, addr
, pte
);
44 WARN_ON(!pte_none(ptent
) && !pte_present(ptent
));
45 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
48 static void vunmap_pmd_range(pud_t
*pud
, unsigned long addr
, unsigned long end
)
53 pmd
= pmd_offset(pud
, addr
);
55 next
= pmd_addr_end(addr
, end
);
56 if (pmd_none_or_clear_bad(pmd
))
58 vunmap_pte_range(pmd
, addr
, next
);
59 } while (pmd
++, addr
= next
, addr
!= end
);
62 static void vunmap_pud_range(pgd_t
*pgd
, unsigned long addr
, unsigned long end
)
67 pud
= pud_offset(pgd
, addr
);
69 next
= pud_addr_end(addr
, end
);
70 if (pud_none_or_clear_bad(pud
))
72 vunmap_pmd_range(pud
, addr
, next
);
73 } while (pud
++, addr
= next
, addr
!= end
);
76 static void vunmap_page_range(unsigned long addr
, unsigned long end
)
82 pgd
= pgd_offset_k(addr
);
84 next
= pgd_addr_end(addr
, end
);
85 if (pgd_none_or_clear_bad(pgd
))
87 vunmap_pud_range(pgd
, addr
, next
);
88 } while (pgd
++, addr
= next
, addr
!= end
);
91 static int vmap_pte_range(pmd_t
*pmd
, unsigned long addr
,
92 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
97 * nr is a running index into the array which helps higher level
98 * callers keep track of where we're up to.
101 pte
= pte_alloc_kernel(pmd
, addr
);
105 struct page
*page
= pages
[*nr
];
107 if (WARN_ON(!pte_none(*pte
)))
111 set_pte_at(&init_mm
, addr
, pte
, mk_pte(page
, prot
));
113 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
117 static int vmap_pmd_range(pud_t
*pud
, unsigned long addr
,
118 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
123 pmd
= pmd_alloc(&init_mm
, pud
, addr
);
127 next
= pmd_addr_end(addr
, end
);
128 if (vmap_pte_range(pmd
, addr
, next
, prot
, pages
, nr
))
130 } while (pmd
++, addr
= next
, addr
!= end
);
134 static int vmap_pud_range(pgd_t
*pgd
, unsigned long addr
,
135 unsigned long end
, pgprot_t prot
, struct page
**pages
, int *nr
)
140 pud
= pud_alloc(&init_mm
, pgd
, addr
);
144 next
= pud_addr_end(addr
, end
);
145 if (vmap_pmd_range(pud
, addr
, next
, prot
, pages
, nr
))
147 } while (pud
++, addr
= next
, addr
!= end
);
152 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
153 * will have pfns corresponding to the "pages" array.
155 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
157 static int vmap_page_range_noflush(unsigned long start
, unsigned long end
,
158 pgprot_t prot
, struct page
**pages
)
162 unsigned long addr
= start
;
167 pgd
= pgd_offset_k(addr
);
169 next
= pgd_addr_end(addr
, end
);
170 err
= vmap_pud_range(pgd
, addr
, next
, prot
, pages
, &nr
);
173 } while (pgd
++, addr
= next
, addr
!= end
);
178 static int vmap_page_range(unsigned long start
, unsigned long end
,
179 pgprot_t prot
, struct page
**pages
)
183 ret
= vmap_page_range_noflush(start
, end
, prot
, pages
);
184 flush_cache_vmap(start
, end
);
188 int is_vmalloc_or_module_addr(const void *x
)
191 * ARM, x86-64 and sparc64 put modules in a special place,
192 * and fall back on vmalloc() if that fails. Others
193 * just put it in the vmalloc space.
195 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
196 unsigned long addr
= (unsigned long)x
;
197 if (addr
>= MODULES_VADDR
&& addr
< MODULES_END
)
200 return is_vmalloc_addr(x
);
204 * Walk a vmap address to the struct page it maps.
206 struct page
*vmalloc_to_page(const void *vmalloc_addr
)
208 unsigned long addr
= (unsigned long) vmalloc_addr
;
209 struct page
*page
= NULL
;
210 pgd_t
*pgd
= pgd_offset_k(addr
);
213 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
214 * architectures that do not vmalloc module space
216 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr
));
218 if (!pgd_none(*pgd
)) {
219 pud_t
*pud
= pud_offset(pgd
, addr
);
220 if (!pud_none(*pud
)) {
221 pmd_t
*pmd
= pmd_offset(pud
, addr
);
222 if (!pmd_none(*pmd
)) {
225 ptep
= pte_offset_map(pmd
, addr
);
227 if (pte_present(pte
))
228 page
= pte_page(pte
);
235 EXPORT_SYMBOL(vmalloc_to_page
);
238 * Map a vmalloc()-space virtual address to the physical page frame number.
240 unsigned long vmalloc_to_pfn(const void *vmalloc_addr
)
242 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
244 EXPORT_SYMBOL(vmalloc_to_pfn
);
247 /*** Global kva allocator ***/
249 #define VM_LAZY_FREE 0x01
250 #define VM_LAZY_FREEING 0x02
251 #define VM_VM_AREA 0x04
254 unsigned long va_start
;
255 unsigned long va_end
;
257 struct rb_node rb_node
; /* address sorted rbtree */
258 struct list_head list
; /* address sorted list */
259 struct list_head purge_list
; /* "lazy purge" list */
261 struct rcu_head rcu_head
;
264 static DEFINE_SPINLOCK(vmap_area_lock
);
265 static struct rb_root vmap_area_root
= RB_ROOT
;
266 static LIST_HEAD(vmap_area_list
);
267 static unsigned long vmap_area_pcpu_hole
;
269 static struct vmap_area
*__find_vmap_area(unsigned long addr
)
271 struct rb_node
*n
= vmap_area_root
.rb_node
;
274 struct vmap_area
*va
;
276 va
= rb_entry(n
, struct vmap_area
, rb_node
);
277 if (addr
< va
->va_start
)
279 else if (addr
> va
->va_start
)
288 static void __insert_vmap_area(struct vmap_area
*va
)
290 struct rb_node
**p
= &vmap_area_root
.rb_node
;
291 struct rb_node
*parent
= NULL
;
295 struct vmap_area
*tmp
;
298 tmp
= rb_entry(parent
, struct vmap_area
, rb_node
);
299 if (va
->va_start
< tmp
->va_end
)
301 else if (va
->va_end
> tmp
->va_start
)
307 rb_link_node(&va
->rb_node
, parent
, p
);
308 rb_insert_color(&va
->rb_node
, &vmap_area_root
);
310 /* address-sort this list so it is usable like the vmlist */
311 tmp
= rb_prev(&va
->rb_node
);
313 struct vmap_area
*prev
;
314 prev
= rb_entry(tmp
, struct vmap_area
, rb_node
);
315 list_add_rcu(&va
->list
, &prev
->list
);
317 list_add_rcu(&va
->list
, &vmap_area_list
);
320 static void purge_vmap_area_lazy(void);
323 * Allocate a region of KVA of the specified size and alignment, within the
326 static struct vmap_area
*alloc_vmap_area(unsigned long size
,
328 unsigned long vstart
, unsigned long vend
,
329 int node
, gfp_t gfp_mask
)
331 struct vmap_area
*va
;
337 BUG_ON(size
& ~PAGE_MASK
);
339 va
= kmalloc_node(sizeof(struct vmap_area
),
340 gfp_mask
& GFP_RECLAIM_MASK
, node
);
342 return ERR_PTR(-ENOMEM
);
345 addr
= ALIGN(vstart
, align
);
347 spin_lock(&vmap_area_lock
);
348 if (addr
+ size
- 1 < addr
)
351 /* XXX: could have a last_hole cache */
352 n
= vmap_area_root
.rb_node
;
354 struct vmap_area
*first
= NULL
;
357 struct vmap_area
*tmp
;
358 tmp
= rb_entry(n
, struct vmap_area
, rb_node
);
359 if (tmp
->va_end
>= addr
) {
360 if (!first
&& tmp
->va_start
< addr
+ size
)
372 if (first
->va_end
< addr
) {
373 n
= rb_next(&first
->rb_node
);
375 first
= rb_entry(n
, struct vmap_area
, rb_node
);
380 while (addr
+ size
> first
->va_start
&& addr
+ size
<= vend
) {
381 addr
= ALIGN(first
->va_end
+ PAGE_SIZE
, align
);
382 if (addr
+ size
- 1 < addr
)
385 n
= rb_next(&first
->rb_node
);
387 first
= rb_entry(n
, struct vmap_area
, rb_node
);
393 if (addr
+ size
> vend
) {
395 spin_unlock(&vmap_area_lock
);
397 purge_vmap_area_lazy();
401 if (printk_ratelimit())
403 "vmap allocation for size %lu failed: "
404 "use vmalloc=<size> to increase size.\n", size
);
406 return ERR_PTR(-EBUSY
);
409 BUG_ON(addr
& (align
-1));
412 va
->va_end
= addr
+ size
;
414 __insert_vmap_area(va
);
415 spin_unlock(&vmap_area_lock
);
420 static void rcu_free_va(struct rcu_head
*head
)
422 struct vmap_area
*va
= container_of(head
, struct vmap_area
, rcu_head
);
427 static void __free_vmap_area(struct vmap_area
*va
)
429 BUG_ON(RB_EMPTY_NODE(&va
->rb_node
));
430 rb_erase(&va
->rb_node
, &vmap_area_root
);
431 RB_CLEAR_NODE(&va
->rb_node
);
432 list_del_rcu(&va
->list
);
435 * Track the highest possible candidate for pcpu area
436 * allocation. Areas outside of vmalloc area can be returned
437 * here too, consider only end addresses which fall inside
438 * vmalloc area proper.
440 if (va
->va_end
> VMALLOC_START
&& va
->va_end
<= VMALLOC_END
)
441 vmap_area_pcpu_hole
= max(vmap_area_pcpu_hole
, va
->va_end
);
443 call_rcu(&va
->rcu_head
, rcu_free_va
);
447 * Free a region of KVA allocated by alloc_vmap_area
449 static void free_vmap_area(struct vmap_area
*va
)
451 spin_lock(&vmap_area_lock
);
452 __free_vmap_area(va
);
453 spin_unlock(&vmap_area_lock
);
457 * Clear the pagetable entries of a given vmap_area
459 static void unmap_vmap_area(struct vmap_area
*va
)
461 vunmap_page_range(va
->va_start
, va
->va_end
);
464 static void vmap_debug_free_range(unsigned long start
, unsigned long end
)
467 * Unmap page tables and force a TLB flush immediately if
468 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
469 * bugs similarly to those in linear kernel virtual address
470 * space after a page has been freed.
472 * All the lazy freeing logic is still retained, in order to
473 * minimise intrusiveness of this debugging feature.
475 * This is going to be *slow* (linear kernel virtual address
476 * debugging doesn't do a broadcast TLB flush so it is a lot
479 #ifdef CONFIG_DEBUG_PAGEALLOC
480 vunmap_page_range(start
, end
);
481 flush_tlb_kernel_range(start
, end
);
486 * lazy_max_pages is the maximum amount of virtual address space we gather up
487 * before attempting to purge with a TLB flush.
489 * There is a tradeoff here: a larger number will cover more kernel page tables
490 * and take slightly longer to purge, but it will linearly reduce the number of
491 * global TLB flushes that must be performed. It would seem natural to scale
492 * this number up linearly with the number of CPUs (because vmapping activity
493 * could also scale linearly with the number of CPUs), however it is likely
494 * that in practice, workloads might be constrained in other ways that mean
495 * vmap activity will not scale linearly with CPUs. Also, I want to be
496 * conservative and not introduce a big latency on huge systems, so go with
497 * a less aggressive log scale. It will still be an improvement over the old
498 * code, and it will be simple to change the scale factor if we find that it
499 * becomes a problem on bigger systems.
501 static unsigned long lazy_max_pages(void)
505 log
= fls(num_online_cpus());
507 return log
* (32UL * 1024 * 1024 / PAGE_SIZE
);
510 static atomic_t vmap_lazy_nr
= ATOMIC_INIT(0);
513 * Purges all lazily-freed vmap areas.
515 * If sync is 0 then don't purge if there is already a purge in progress.
516 * If force_flush is 1, then flush kernel TLBs between *start and *end even
517 * if we found no lazy vmap areas to unmap (callers can use this to optimise
518 * their own TLB flushing).
519 * Returns with *start = min(*start, lowest purged address)
520 * *end = max(*end, highest purged address)
522 static void __purge_vmap_area_lazy(unsigned long *start
, unsigned long *end
,
523 int sync
, int force_flush
)
525 static DEFINE_SPINLOCK(purge_lock
);
527 struct vmap_area
*va
;
528 struct vmap_area
*n_va
;
532 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
533 * should not expect such behaviour. This just simplifies locking for
534 * the case that isn't actually used at the moment anyway.
536 if (!sync
&& !force_flush
) {
537 if (!spin_trylock(&purge_lock
))
540 spin_lock(&purge_lock
);
543 list_for_each_entry_rcu(va
, &vmap_area_list
, list
) {
544 if (va
->flags
& VM_LAZY_FREE
) {
545 if (va
->va_start
< *start
)
546 *start
= va
->va_start
;
547 if (va
->va_end
> *end
)
549 nr
+= (va
->va_end
- va
->va_start
) >> PAGE_SHIFT
;
551 list_add_tail(&va
->purge_list
, &valist
);
552 va
->flags
|= VM_LAZY_FREEING
;
553 va
->flags
&= ~VM_LAZY_FREE
;
559 BUG_ON(nr
> atomic_read(&vmap_lazy_nr
));
560 atomic_sub(nr
, &vmap_lazy_nr
);
563 if (nr
|| force_flush
)
564 flush_tlb_kernel_range(*start
, *end
);
567 spin_lock(&vmap_area_lock
);
568 list_for_each_entry_safe(va
, n_va
, &valist
, purge_list
)
569 __free_vmap_area(va
);
570 spin_unlock(&vmap_area_lock
);
572 spin_unlock(&purge_lock
);
576 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
577 * is already purging.
579 static void try_purge_vmap_area_lazy(void)
581 unsigned long start
= ULONG_MAX
, end
= 0;
583 __purge_vmap_area_lazy(&start
, &end
, 0, 0);
587 * Kick off a purge of the outstanding lazy areas.
589 static void purge_vmap_area_lazy(void)
591 unsigned long start
= ULONG_MAX
, end
= 0;
593 __purge_vmap_area_lazy(&start
, &end
, 1, 0);
597 * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
598 * called for the correct range previously.
600 static void free_unmap_vmap_area_noflush(struct vmap_area
*va
)
602 va
->flags
|= VM_LAZY_FREE
;
603 atomic_add((va
->va_end
- va
->va_start
) >> PAGE_SHIFT
, &vmap_lazy_nr
);
604 if (unlikely(atomic_read(&vmap_lazy_nr
) > lazy_max_pages()))
605 try_purge_vmap_area_lazy();
609 * Free and unmap a vmap area
611 static void free_unmap_vmap_area(struct vmap_area
*va
)
613 flush_cache_vunmap(va
->va_start
, va
->va_end
);
614 free_unmap_vmap_area_noflush(va
);
617 static struct vmap_area
*find_vmap_area(unsigned long addr
)
619 struct vmap_area
*va
;
621 spin_lock(&vmap_area_lock
);
622 va
= __find_vmap_area(addr
);
623 spin_unlock(&vmap_area_lock
);
628 static void free_unmap_vmap_area_addr(unsigned long addr
)
630 struct vmap_area
*va
;
632 va
= find_vmap_area(addr
);
634 free_unmap_vmap_area(va
);
638 /*** Per cpu kva allocator ***/
641 * vmap space is limited especially on 32 bit architectures. Ensure there is
642 * room for at least 16 percpu vmap blocks per CPU.
645 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
646 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
647 * instead (we just need a rough idea)
649 #if BITS_PER_LONG == 32
650 #define VMALLOC_SPACE (128UL*1024*1024)
652 #define VMALLOC_SPACE (128UL*1024*1024*1024)
655 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
656 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
657 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
658 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
659 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
660 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
661 #define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
662 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
663 VMALLOC_PAGES / NR_CPUS / 16))
665 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
667 static bool vmap_initialized __read_mostly
= false;
669 struct vmap_block_queue
{
671 struct list_head free
;
672 struct list_head dirty
;
673 unsigned int nr_dirty
;
678 struct vmap_area
*va
;
679 struct vmap_block_queue
*vbq
;
680 unsigned long free
, dirty
;
681 DECLARE_BITMAP(alloc_map
, VMAP_BBMAP_BITS
);
682 DECLARE_BITMAP(dirty_map
, VMAP_BBMAP_BITS
);
684 struct list_head free_list
;
685 struct rcu_head rcu_head
;
689 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
690 static DEFINE_PER_CPU(struct vmap_block_queue
, vmap_block_queue
);
693 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
694 * in the free path. Could get rid of this if we change the API to return a
695 * "cookie" from alloc, to be passed to free. But no big deal yet.
697 static DEFINE_SPINLOCK(vmap_block_tree_lock
);
698 static RADIX_TREE(vmap_block_tree
, GFP_ATOMIC
);
701 * We should probably have a fallback mechanism to allocate virtual memory
702 * out of partially filled vmap blocks. However vmap block sizing should be
703 * fairly reasonable according to the vmalloc size, so it shouldn't be a
707 static unsigned long addr_to_vb_idx(unsigned long addr
)
709 addr
-= VMALLOC_START
& ~(VMAP_BLOCK_SIZE
-1);
710 addr
/= VMAP_BLOCK_SIZE
;
714 static struct vmap_block
*new_vmap_block(gfp_t gfp_mask
)
716 struct vmap_block_queue
*vbq
;
717 struct vmap_block
*vb
;
718 struct vmap_area
*va
;
719 unsigned long vb_idx
;
722 node
= numa_node_id();
724 vb
= kmalloc_node(sizeof(struct vmap_block
),
725 gfp_mask
& GFP_RECLAIM_MASK
, node
);
727 return ERR_PTR(-ENOMEM
);
729 va
= alloc_vmap_area(VMAP_BLOCK_SIZE
, VMAP_BLOCK_SIZE
,
730 VMALLOC_START
, VMALLOC_END
,
732 if (unlikely(IS_ERR(va
))) {
734 return ERR_PTR(PTR_ERR(va
));
737 err
= radix_tree_preload(gfp_mask
);
744 spin_lock_init(&vb
->lock
);
746 vb
->free
= VMAP_BBMAP_BITS
;
748 bitmap_zero(vb
->alloc_map
, VMAP_BBMAP_BITS
);
749 bitmap_zero(vb
->dirty_map
, VMAP_BBMAP_BITS
);
750 INIT_LIST_HEAD(&vb
->free_list
);
752 vb_idx
= addr_to_vb_idx(va
->va_start
);
753 spin_lock(&vmap_block_tree_lock
);
754 err
= radix_tree_insert(&vmap_block_tree
, vb_idx
, vb
);
755 spin_unlock(&vmap_block_tree_lock
);
757 radix_tree_preload_end();
759 vbq
= &get_cpu_var(vmap_block_queue
);
761 spin_lock(&vbq
->lock
);
762 list_add(&vb
->free_list
, &vbq
->free
);
763 spin_unlock(&vbq
->lock
);
764 put_cpu_var(vmap_cpu_blocks
);
769 static void rcu_free_vb(struct rcu_head
*head
)
771 struct vmap_block
*vb
= container_of(head
, struct vmap_block
, rcu_head
);
776 static void free_vmap_block(struct vmap_block
*vb
)
778 struct vmap_block
*tmp
;
779 unsigned long vb_idx
;
781 BUG_ON(!list_empty(&vb
->free_list
));
783 vb_idx
= addr_to_vb_idx(vb
->va
->va_start
);
784 spin_lock(&vmap_block_tree_lock
);
785 tmp
= radix_tree_delete(&vmap_block_tree
, vb_idx
);
786 spin_unlock(&vmap_block_tree_lock
);
789 free_unmap_vmap_area_noflush(vb
->va
);
790 call_rcu(&vb
->rcu_head
, rcu_free_vb
);
793 static void *vb_alloc(unsigned long size
, gfp_t gfp_mask
)
795 struct vmap_block_queue
*vbq
;
796 struct vmap_block
*vb
;
797 unsigned long addr
= 0;
800 BUG_ON(size
& ~PAGE_MASK
);
801 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
802 order
= get_order(size
);
806 vbq
= &get_cpu_var(vmap_block_queue
);
807 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
810 spin_lock(&vb
->lock
);
811 i
= bitmap_find_free_region(vb
->alloc_map
,
812 VMAP_BBMAP_BITS
, order
);
815 addr
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
816 BUG_ON(addr_to_vb_idx(addr
) !=
817 addr_to_vb_idx(vb
->va
->va_start
));
818 vb
->free
-= 1UL << order
;
820 spin_lock(&vbq
->lock
);
821 list_del_init(&vb
->free_list
);
822 spin_unlock(&vbq
->lock
);
824 spin_unlock(&vb
->lock
);
827 spin_unlock(&vb
->lock
);
829 put_cpu_var(vmap_cpu_blocks
);
833 vb
= new_vmap_block(gfp_mask
);
842 static void vb_free(const void *addr
, unsigned long size
)
844 unsigned long offset
;
845 unsigned long vb_idx
;
847 struct vmap_block
*vb
;
849 BUG_ON(size
& ~PAGE_MASK
);
850 BUG_ON(size
> PAGE_SIZE
*VMAP_MAX_ALLOC
);
852 flush_cache_vunmap((unsigned long)addr
, (unsigned long)addr
+ size
);
854 order
= get_order(size
);
856 offset
= (unsigned long)addr
& (VMAP_BLOCK_SIZE
- 1);
858 vb_idx
= addr_to_vb_idx((unsigned long)addr
);
860 vb
= radix_tree_lookup(&vmap_block_tree
, vb_idx
);
864 spin_lock(&vb
->lock
);
865 bitmap_allocate_region(vb
->dirty_map
, offset
>> PAGE_SHIFT
, order
);
867 vb
->dirty
+= 1UL << order
;
868 if (vb
->dirty
== VMAP_BBMAP_BITS
) {
869 BUG_ON(vb
->free
|| !list_empty(&vb
->free_list
));
870 spin_unlock(&vb
->lock
);
873 spin_unlock(&vb
->lock
);
877 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
879 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
880 * to amortize TLB flushing overheads. What this means is that any page you
881 * have now, may, in a former life, have been mapped into kernel virtual
882 * address by the vmap layer and so there might be some CPUs with TLB entries
883 * still referencing that page (additional to the regular 1:1 kernel mapping).
885 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
886 * be sure that none of the pages we have control over will have any aliases
887 * from the vmap layer.
889 void vm_unmap_aliases(void)
891 unsigned long start
= ULONG_MAX
, end
= 0;
895 if (unlikely(!vmap_initialized
))
898 for_each_possible_cpu(cpu
) {
899 struct vmap_block_queue
*vbq
= &per_cpu(vmap_block_queue
, cpu
);
900 struct vmap_block
*vb
;
903 list_for_each_entry_rcu(vb
, &vbq
->free
, free_list
) {
906 spin_lock(&vb
->lock
);
907 i
= find_first_bit(vb
->dirty_map
, VMAP_BBMAP_BITS
);
908 while (i
< VMAP_BBMAP_BITS
) {
911 j
= find_next_zero_bit(vb
->dirty_map
,
914 s
= vb
->va
->va_start
+ (i
<< PAGE_SHIFT
);
915 e
= vb
->va
->va_start
+ (j
<< PAGE_SHIFT
);
916 vunmap_page_range(s
, e
);
925 i
= find_next_bit(vb
->dirty_map
,
928 spin_unlock(&vb
->lock
);
933 __purge_vmap_area_lazy(&start
, &end
, 1, flush
);
935 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
938 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
939 * @mem: the pointer returned by vm_map_ram
940 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
942 void vm_unmap_ram(const void *mem
, unsigned int count
)
944 unsigned long size
= count
<< PAGE_SHIFT
;
945 unsigned long addr
= (unsigned long)mem
;
948 BUG_ON(addr
< VMALLOC_START
);
949 BUG_ON(addr
> VMALLOC_END
);
950 BUG_ON(addr
& (PAGE_SIZE
-1));
952 debug_check_no_locks_freed(mem
, size
);
953 vmap_debug_free_range(addr
, addr
+size
);
955 if (likely(count
<= VMAP_MAX_ALLOC
))
958 free_unmap_vmap_area_addr(addr
);
960 EXPORT_SYMBOL(vm_unmap_ram
);
963 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
964 * @pages: an array of pointers to the pages to be mapped
965 * @count: number of pages
966 * @node: prefer to allocate data structures on this node
967 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
969 * Returns: a pointer to the address that has been mapped, or %NULL on failure
971 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
973 unsigned long size
= count
<< PAGE_SHIFT
;
977 if (likely(count
<= VMAP_MAX_ALLOC
)) {
978 mem
= vb_alloc(size
, GFP_KERNEL
);
981 addr
= (unsigned long)mem
;
983 struct vmap_area
*va
;
984 va
= alloc_vmap_area(size
, PAGE_SIZE
,
985 VMALLOC_START
, VMALLOC_END
, node
, GFP_KERNEL
);
992 if (vmap_page_range(addr
, addr
+ size
, prot
, pages
) < 0) {
993 vm_unmap_ram(mem
, count
);
998 EXPORT_SYMBOL(vm_map_ram
);
1001 * vm_area_register_early - register vmap area early during boot
1002 * @vm: vm_struct to register
1003 * @align: requested alignment
1005 * This function is used to register kernel vm area before
1006 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1007 * proper values on entry and other fields should be zero. On return,
1008 * vm->addr contains the allocated address.
1010 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1012 void __init
vm_area_register_early(struct vm_struct
*vm
, size_t align
)
1014 static size_t vm_init_off __initdata
;
1017 addr
= ALIGN(VMALLOC_START
+ vm_init_off
, align
);
1018 vm_init_off
= PFN_ALIGN(addr
+ vm
->size
) - VMALLOC_START
;
1020 vm
->addr
= (void *)addr
;
1026 void __init
vmalloc_init(void)
1028 struct vmap_area
*va
;
1029 struct vm_struct
*tmp
;
1032 for_each_possible_cpu(i
) {
1033 struct vmap_block_queue
*vbq
;
1035 vbq
= &per_cpu(vmap_block_queue
, i
);
1036 spin_lock_init(&vbq
->lock
);
1037 INIT_LIST_HEAD(&vbq
->free
);
1038 INIT_LIST_HEAD(&vbq
->dirty
);
1042 /* Import existing vmlist entries. */
1043 for (tmp
= vmlist
; tmp
; tmp
= tmp
->next
) {
1044 va
= kzalloc(sizeof(struct vmap_area
), GFP_NOWAIT
);
1045 va
->flags
= tmp
->flags
| VM_VM_AREA
;
1046 va
->va_start
= (unsigned long)tmp
->addr
;
1047 va
->va_end
= va
->va_start
+ tmp
->size
;
1048 __insert_vmap_area(va
);
1051 vmap_area_pcpu_hole
= VMALLOC_END
;
1053 vmap_initialized
= true;
1057 * map_kernel_range_noflush - map kernel VM area with the specified pages
1058 * @addr: start of the VM area to map
1059 * @size: size of the VM area to map
1060 * @prot: page protection flags to use
1061 * @pages: pages to map
1063 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1064 * specify should have been allocated using get_vm_area() and its
1068 * This function does NOT do any cache flushing. The caller is
1069 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1070 * before calling this function.
1073 * The number of pages mapped on success, -errno on failure.
1075 int map_kernel_range_noflush(unsigned long addr
, unsigned long size
,
1076 pgprot_t prot
, struct page
**pages
)
1078 return vmap_page_range_noflush(addr
, addr
+ size
, prot
, pages
);
1082 * unmap_kernel_range_noflush - unmap kernel VM area
1083 * @addr: start of the VM area to unmap
1084 * @size: size of the VM area to unmap
1086 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1087 * specify should have been allocated using get_vm_area() and its
1091 * This function does NOT do any cache flushing. The caller is
1092 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1093 * before calling this function and flush_tlb_kernel_range() after.
1095 void unmap_kernel_range_noflush(unsigned long addr
, unsigned long size
)
1097 vunmap_page_range(addr
, addr
+ size
);
1101 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1102 * @addr: start of the VM area to unmap
1103 * @size: size of the VM area to unmap
1105 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1106 * the unmapping and tlb after.
1108 void unmap_kernel_range(unsigned long addr
, unsigned long size
)
1110 unsigned long end
= addr
+ size
;
1112 flush_cache_vunmap(addr
, end
);
1113 vunmap_page_range(addr
, end
);
1114 flush_tlb_kernel_range(addr
, end
);
1117 int map_vm_area(struct vm_struct
*area
, pgprot_t prot
, struct page
***pages
)
1119 unsigned long addr
= (unsigned long)area
->addr
;
1120 unsigned long end
= addr
+ area
->size
- PAGE_SIZE
;
1123 err
= vmap_page_range(addr
, end
, prot
, *pages
);
1131 EXPORT_SYMBOL_GPL(map_vm_area
);
1133 /*** Old vmalloc interfaces ***/
1134 DEFINE_RWLOCK(vmlist_lock
);
1135 struct vm_struct
*vmlist
;
1137 static void insert_vmalloc_vm(struct vm_struct
*vm
, struct vmap_area
*va
,
1138 unsigned long flags
, void *caller
)
1140 struct vm_struct
*tmp
, **p
;
1143 vm
->addr
= (void *)va
->va_start
;
1144 vm
->size
= va
->va_end
- va
->va_start
;
1145 vm
->caller
= caller
;
1147 va
->flags
|= VM_VM_AREA
;
1149 write_lock(&vmlist_lock
);
1150 for (p
= &vmlist
; (tmp
= *p
) != NULL
; p
= &tmp
->next
) {
1151 if (tmp
->addr
>= vm
->addr
)
1156 write_unlock(&vmlist_lock
);
1159 static struct vm_struct
*__get_vm_area_node(unsigned long size
,
1160 unsigned long align
, unsigned long flags
, unsigned long start
,
1161 unsigned long end
, int node
, gfp_t gfp_mask
, void *caller
)
1163 static struct vmap_area
*va
;
1164 struct vm_struct
*area
;
1166 BUG_ON(in_interrupt());
1167 if (flags
& VM_IOREMAP
) {
1168 int bit
= fls(size
);
1170 if (bit
> IOREMAP_MAX_ORDER
)
1171 bit
= IOREMAP_MAX_ORDER
;
1172 else if (bit
< PAGE_SHIFT
)
1178 size
= PAGE_ALIGN(size
);
1179 if (unlikely(!size
))
1182 area
= kzalloc_node(sizeof(*area
), gfp_mask
& GFP_RECLAIM_MASK
, node
);
1183 if (unlikely(!area
))
1187 * We always allocate a guard page.
1191 va
= alloc_vmap_area(size
, align
, start
, end
, node
, gfp_mask
);
1197 insert_vmalloc_vm(area
, va
, flags
, caller
);
1201 struct vm_struct
*__get_vm_area(unsigned long size
, unsigned long flags
,
1202 unsigned long start
, unsigned long end
)
1204 return __get_vm_area_node(size
, 1, flags
, start
, end
, -1, GFP_KERNEL
,
1205 __builtin_return_address(0));
1207 EXPORT_SYMBOL_GPL(__get_vm_area
);
1209 struct vm_struct
*__get_vm_area_caller(unsigned long size
, unsigned long flags
,
1210 unsigned long start
, unsigned long end
,
1213 return __get_vm_area_node(size
, 1, flags
, start
, end
, -1, GFP_KERNEL
,
1218 * get_vm_area - reserve a contiguous kernel virtual area
1219 * @size: size of the area
1220 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1222 * Search an area of @size in the kernel virtual mapping area,
1223 * and reserved it for out purposes. Returns the area descriptor
1224 * on success or %NULL on failure.
1226 struct vm_struct
*get_vm_area(unsigned long size
, unsigned long flags
)
1228 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1229 -1, GFP_KERNEL
, __builtin_return_address(0));
1232 struct vm_struct
*get_vm_area_caller(unsigned long size
, unsigned long flags
,
1235 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1236 -1, GFP_KERNEL
, caller
);
1239 struct vm_struct
*get_vm_area_node(unsigned long size
, unsigned long flags
,
1240 int node
, gfp_t gfp_mask
)
1242 return __get_vm_area_node(size
, 1, flags
, VMALLOC_START
, VMALLOC_END
,
1243 node
, gfp_mask
, __builtin_return_address(0));
1246 static struct vm_struct
*find_vm_area(const void *addr
)
1248 struct vmap_area
*va
;
1250 va
= find_vmap_area((unsigned long)addr
);
1251 if (va
&& va
->flags
& VM_VM_AREA
)
1258 * remove_vm_area - find and remove a continuous kernel virtual area
1259 * @addr: base address
1261 * Search for the kernel VM area starting at @addr, and remove it.
1262 * This function returns the found VM area, but using it is NOT safe
1263 * on SMP machines, except for its size or flags.
1265 struct vm_struct
*remove_vm_area(const void *addr
)
1267 struct vmap_area
*va
;
1269 va
= find_vmap_area((unsigned long)addr
);
1270 if (va
&& va
->flags
& VM_VM_AREA
) {
1271 struct vm_struct
*vm
= va
->private;
1272 struct vm_struct
*tmp
, **p
;
1274 * remove from list and disallow access to this vm_struct
1275 * before unmap. (address range confliction is maintained by
1278 write_lock(&vmlist_lock
);
1279 for (p
= &vmlist
; (tmp
= *p
) != vm
; p
= &tmp
->next
)
1282 write_unlock(&vmlist_lock
);
1284 vmap_debug_free_range(va
->va_start
, va
->va_end
);
1285 free_unmap_vmap_area(va
);
1286 vm
->size
-= PAGE_SIZE
;
1293 static void __vunmap(const void *addr
, int deallocate_pages
)
1295 struct vm_struct
*area
;
1300 if ((PAGE_SIZE
-1) & (unsigned long)addr
) {
1301 WARN(1, KERN_ERR
"Trying to vfree() bad address (%p)\n", addr
);
1305 area
= remove_vm_area(addr
);
1306 if (unlikely(!area
)) {
1307 WARN(1, KERN_ERR
"Trying to vfree() nonexistent vm area (%p)\n",
1312 debug_check_no_locks_freed(addr
, area
->size
);
1313 debug_check_no_obj_freed(addr
, area
->size
);
1315 if (deallocate_pages
) {
1318 for (i
= 0; i
< area
->nr_pages
; i
++) {
1319 struct page
*page
= area
->pages
[i
];
1325 if (area
->flags
& VM_VPAGES
)
1336 * vfree - release memory allocated by vmalloc()
1337 * @addr: memory base address
1339 * Free the virtually continuous memory area starting at @addr, as
1340 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1341 * NULL, no operation is performed.
1343 * Must not be called in interrupt context.
1345 void vfree(const void *addr
)
1347 BUG_ON(in_interrupt());
1349 kmemleak_free(addr
);
1353 EXPORT_SYMBOL(vfree
);
1356 * vunmap - release virtual mapping obtained by vmap()
1357 * @addr: memory base address
1359 * Free the virtually contiguous memory area starting at @addr,
1360 * which was created from the page array passed to vmap().
1362 * Must not be called in interrupt context.
1364 void vunmap(const void *addr
)
1366 BUG_ON(in_interrupt());
1370 EXPORT_SYMBOL(vunmap
);
1373 * vmap - map an array of pages into virtually contiguous space
1374 * @pages: array of page pointers
1375 * @count: number of pages to map
1376 * @flags: vm_area->flags
1377 * @prot: page protection for the mapping
1379 * Maps @count pages from @pages into contiguous kernel virtual
1382 void *vmap(struct page
**pages
, unsigned int count
,
1383 unsigned long flags
, pgprot_t prot
)
1385 struct vm_struct
*area
;
1389 if (count
> totalram_pages
)
1392 area
= get_vm_area_caller((count
<< PAGE_SHIFT
), flags
,
1393 __builtin_return_address(0));
1397 if (map_vm_area(area
, prot
, &pages
)) {
1404 EXPORT_SYMBOL(vmap
);
1406 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1407 gfp_t gfp_mask
, pgprot_t prot
,
1408 int node
, void *caller
);
1409 static void *__vmalloc_area_node(struct vm_struct
*area
, gfp_t gfp_mask
,
1410 pgprot_t prot
, int node
, void *caller
)
1412 struct page
**pages
;
1413 unsigned int nr_pages
, array_size
, i
;
1414 gfp_t nested_gfp
= (gfp_mask
& GFP_RECLAIM_MASK
) | __GFP_ZERO
;
1416 nr_pages
= (area
->size
- PAGE_SIZE
) >> PAGE_SHIFT
;
1417 array_size
= (nr_pages
* sizeof(struct page
*));
1419 area
->nr_pages
= nr_pages
;
1420 /* Please note that the recursion is strictly bounded. */
1421 if (array_size
> PAGE_SIZE
) {
1422 pages
= __vmalloc_node(array_size
, 1, nested_gfp
|__GFP_HIGHMEM
,
1423 PAGE_KERNEL
, node
, caller
);
1424 area
->flags
|= VM_VPAGES
;
1426 pages
= kmalloc_node(array_size
, nested_gfp
, node
);
1428 area
->pages
= pages
;
1429 area
->caller
= caller
;
1431 remove_vm_area(area
->addr
);
1436 for (i
= 0; i
< area
->nr_pages
; i
++) {
1440 page
= alloc_page(gfp_mask
);
1442 page
= alloc_pages_node(node
, gfp_mask
, 0);
1444 if (unlikely(!page
)) {
1445 /* Successfully allocated i pages, free them in __vunmap() */
1449 area
->pages
[i
] = page
;
1452 if (map_vm_area(area
, prot
, &pages
))
1461 void *__vmalloc_area(struct vm_struct
*area
, gfp_t gfp_mask
, pgprot_t prot
)
1463 void *addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, -1,
1464 __builtin_return_address(0));
1467 * A ref_count = 3 is needed because the vm_struct and vmap_area
1468 * structures allocated in the __get_vm_area_node() function contain
1469 * references to the virtual address of the vmalloc'ed block.
1471 kmemleak_alloc(addr
, area
->size
- PAGE_SIZE
, 3, gfp_mask
);
1477 * __vmalloc_node - allocate virtually contiguous memory
1478 * @size: allocation size
1479 * @align: desired alignment
1480 * @gfp_mask: flags for the page level allocator
1481 * @prot: protection mask for the allocated pages
1482 * @node: node to use for allocation or -1
1483 * @caller: caller's return address
1485 * Allocate enough pages to cover @size from the page level
1486 * allocator with @gfp_mask flags. Map them into contiguous
1487 * kernel virtual space, using a pagetable protection of @prot.
1489 static void *__vmalloc_node(unsigned long size
, unsigned long align
,
1490 gfp_t gfp_mask
, pgprot_t prot
,
1491 int node
, void *caller
)
1493 struct vm_struct
*area
;
1495 unsigned long real_size
= size
;
1497 size
= PAGE_ALIGN(size
);
1498 if (!size
|| (size
>> PAGE_SHIFT
) > totalram_pages
)
1501 area
= __get_vm_area_node(size
, align
, VM_ALLOC
, VMALLOC_START
,
1502 VMALLOC_END
, node
, gfp_mask
, caller
);
1507 addr
= __vmalloc_area_node(area
, gfp_mask
, prot
, node
, caller
);
1510 * A ref_count = 3 is needed because the vm_struct and vmap_area
1511 * structures allocated in the __get_vm_area_node() function contain
1512 * references to the virtual address of the vmalloc'ed block.
1514 kmemleak_alloc(addr
, real_size
, 3, gfp_mask
);
1519 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
1521 return __vmalloc_node(size
, 1, gfp_mask
, prot
, -1,
1522 __builtin_return_address(0));
1524 EXPORT_SYMBOL(__vmalloc
);
1527 * vmalloc - allocate virtually contiguous memory
1528 * @size: allocation size
1529 * Allocate enough pages to cover @size from the page level
1530 * allocator and map them into contiguous kernel virtual space.
1532 * For tight control over page level allocator and protection flags
1533 * use __vmalloc() instead.
1535 void *vmalloc(unsigned long size
)
1537 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1538 -1, __builtin_return_address(0));
1540 EXPORT_SYMBOL(vmalloc
);
1543 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1544 * @size: allocation size
1546 * The resulting memory area is zeroed so it can be mapped to userspace
1547 * without leaking data.
1549 void *vmalloc_user(unsigned long size
)
1551 struct vm_struct
*area
;
1554 ret
= __vmalloc_node(size
, SHMLBA
,
1555 GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
1556 PAGE_KERNEL
, -1, __builtin_return_address(0));
1558 area
= find_vm_area(ret
);
1559 area
->flags
|= VM_USERMAP
;
1563 EXPORT_SYMBOL(vmalloc_user
);
1566 * vmalloc_node - allocate memory on a specific node
1567 * @size: allocation size
1570 * Allocate enough pages to cover @size from the page level
1571 * allocator and map them into contiguous kernel virtual space.
1573 * For tight control over page level allocator and protection flags
1574 * use __vmalloc() instead.
1576 void *vmalloc_node(unsigned long size
, int node
)
1578 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
,
1579 node
, __builtin_return_address(0));
1581 EXPORT_SYMBOL(vmalloc_node
);
1583 #ifndef PAGE_KERNEL_EXEC
1584 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1588 * vmalloc_exec - allocate virtually contiguous, executable memory
1589 * @size: allocation size
1591 * Kernel-internal function to allocate enough pages to cover @size
1592 * the page level allocator and map them into contiguous and
1593 * executable kernel virtual space.
1595 * For tight control over page level allocator and protection flags
1596 * use __vmalloc() instead.
1599 void *vmalloc_exec(unsigned long size
)
1601 return __vmalloc_node(size
, 1, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
,
1602 -1, __builtin_return_address(0));
1605 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1606 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1607 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1608 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1610 #define GFP_VMALLOC32 GFP_KERNEL
1614 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1615 * @size: allocation size
1617 * Allocate enough 32bit PA addressable pages to cover @size from the
1618 * page level allocator and map them into contiguous kernel virtual space.
1620 void *vmalloc_32(unsigned long size
)
1622 return __vmalloc_node(size
, 1, GFP_VMALLOC32
, PAGE_KERNEL
,
1623 -1, __builtin_return_address(0));
1625 EXPORT_SYMBOL(vmalloc_32
);
1628 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1629 * @size: allocation size
1631 * The resulting memory area is 32bit addressable and zeroed so it can be
1632 * mapped to userspace without leaking data.
1634 void *vmalloc_32_user(unsigned long size
)
1636 struct vm_struct
*area
;
1639 ret
= __vmalloc_node(size
, 1, GFP_VMALLOC32
| __GFP_ZERO
, PAGE_KERNEL
,
1640 -1, __builtin_return_address(0));
1642 area
= find_vm_area(ret
);
1643 area
->flags
|= VM_USERMAP
;
1647 EXPORT_SYMBOL(vmalloc_32_user
);
1650 * small helper routine , copy contents to buf from addr.
1651 * If the page is not present, fill zero.
1654 static int aligned_vread(char *buf
, char *addr
, unsigned long count
)
1660 unsigned long offset
, length
;
1662 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1663 length
= PAGE_SIZE
- offset
;
1666 p
= vmalloc_to_page(addr
);
1668 * To do safe access to this _mapped_ area, we need
1669 * lock. But adding lock here means that we need to add
1670 * overhead of vmalloc()/vfree() calles for this _debug_
1671 * interface, rarely used. Instead of that, we'll use
1672 * kmap() and get small overhead in this access function.
1676 * we can expect USER0 is not used (see vread/vwrite's
1677 * function description)
1679 void *map
= kmap_atomic(p
, KM_USER0
);
1680 memcpy(buf
, map
+ offset
, length
);
1681 kunmap_atomic(map
, KM_USER0
);
1683 memset(buf
, 0, length
);
1693 static int aligned_vwrite(char *buf
, char *addr
, unsigned long count
)
1699 unsigned long offset
, length
;
1701 offset
= (unsigned long)addr
& ~PAGE_MASK
;
1702 length
= PAGE_SIZE
- offset
;
1705 p
= vmalloc_to_page(addr
);
1707 * To do safe access to this _mapped_ area, we need
1708 * lock. But adding lock here means that we need to add
1709 * overhead of vmalloc()/vfree() calles for this _debug_
1710 * interface, rarely used. Instead of that, we'll use
1711 * kmap() and get small overhead in this access function.
1715 * we can expect USER0 is not used (see vread/vwrite's
1716 * function description)
1718 void *map
= kmap_atomic(p
, KM_USER0
);
1719 memcpy(map
+ offset
, buf
, length
);
1720 kunmap_atomic(map
, KM_USER0
);
1731 * vread() - read vmalloc area in a safe way.
1732 * @buf: buffer for reading data
1733 * @addr: vm address.
1734 * @count: number of bytes to be read.
1736 * Returns # of bytes which addr and buf should be increased.
1737 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
1738 * includes any intersect with alive vmalloc area.
1740 * This function checks that addr is a valid vmalloc'ed area, and
1741 * copy data from that area to a given buffer. If the given memory range
1742 * of [addr...addr+count) includes some valid address, data is copied to
1743 * proper area of @buf. If there are memory holes, they'll be zero-filled.
1744 * IOREMAP area is treated as memory hole and no copy is done.
1746 * If [addr...addr+count) doesn't includes any intersects with alive
1747 * vm_struct area, returns 0.
1748 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1749 * the caller should guarantee KM_USER0 is not used.
1751 * Note: In usual ops, vread() is never necessary because the caller
1752 * should know vmalloc() area is valid and can use memcpy().
1753 * This is for routines which have to access vmalloc area without
1754 * any informaion, as /dev/kmem.
1758 long vread(char *buf
, char *addr
, unsigned long count
)
1760 struct vm_struct
*tmp
;
1761 char *vaddr
, *buf_start
= buf
;
1762 unsigned long buflen
= count
;
1765 /* Don't allow overflow */
1766 if ((unsigned long) addr
+ count
< count
)
1767 count
= -(unsigned long) addr
;
1769 read_lock(&vmlist_lock
);
1770 for (tmp
= vmlist
; count
&& tmp
; tmp
= tmp
->next
) {
1771 vaddr
= (char *) tmp
->addr
;
1772 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
1774 while (addr
< vaddr
) {
1782 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
1785 if (!(tmp
->flags
& VM_IOREMAP
))
1786 aligned_vread(buf
, addr
, n
);
1787 else /* IOREMAP area is treated as memory hole */
1794 read_unlock(&vmlist_lock
);
1796 if (buf
== buf_start
)
1798 /* zero-fill memory holes */
1799 if (buf
!= buf_start
+ buflen
)
1800 memset(buf
, 0, buflen
- (buf
- buf_start
));
1806 * vwrite() - write vmalloc area in a safe way.
1807 * @buf: buffer for source data
1808 * @addr: vm address.
1809 * @count: number of bytes to be read.
1811 * Returns # of bytes which addr and buf should be incresed.
1812 * (same number to @count).
1813 * If [addr...addr+count) doesn't includes any intersect with valid
1814 * vmalloc area, returns 0.
1816 * This function checks that addr is a valid vmalloc'ed area, and
1817 * copy data from a buffer to the given addr. If specified range of
1818 * [addr...addr+count) includes some valid address, data is copied from
1819 * proper area of @buf. If there are memory holes, no copy to hole.
1820 * IOREMAP area is treated as memory hole and no copy is done.
1822 * If [addr...addr+count) doesn't includes any intersects with alive
1823 * vm_struct area, returns 0.
1824 * @buf should be kernel's buffer. Because this function uses KM_USER0,
1825 * the caller should guarantee KM_USER0 is not used.
1827 * Note: In usual ops, vwrite() is never necessary because the caller
1828 * should know vmalloc() area is valid and can use memcpy().
1829 * This is for routines which have to access vmalloc area without
1830 * any informaion, as /dev/kmem.
1832 * The caller should guarantee KM_USER1 is not used.
1835 long vwrite(char *buf
, char *addr
, unsigned long count
)
1837 struct vm_struct
*tmp
;
1839 unsigned long n
, buflen
;
1842 /* Don't allow overflow */
1843 if ((unsigned long) addr
+ count
< count
)
1844 count
= -(unsigned long) addr
;
1847 read_lock(&vmlist_lock
);
1848 for (tmp
= vmlist
; count
&& tmp
; tmp
= tmp
->next
) {
1849 vaddr
= (char *) tmp
->addr
;
1850 if (addr
>= vaddr
+ tmp
->size
- PAGE_SIZE
)
1852 while (addr
< vaddr
) {
1859 n
= vaddr
+ tmp
->size
- PAGE_SIZE
- addr
;
1862 if (!(tmp
->flags
& VM_IOREMAP
)) {
1863 aligned_vwrite(buf
, addr
, n
);
1871 read_unlock(&vmlist_lock
);
1878 * remap_vmalloc_range - map vmalloc pages to userspace
1879 * @vma: vma to cover (map full range of vma)
1880 * @addr: vmalloc memory
1881 * @pgoff: number of pages into addr before first page to map
1883 * Returns: 0 for success, -Exxx on failure
1885 * This function checks that addr is a valid vmalloc'ed area, and
1886 * that it is big enough to cover the vma. Will return failure if
1887 * that criteria isn't met.
1889 * Similar to remap_pfn_range() (see mm/memory.c)
1891 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1892 unsigned long pgoff
)
1894 struct vm_struct
*area
;
1895 unsigned long uaddr
= vma
->vm_start
;
1896 unsigned long usize
= vma
->vm_end
- vma
->vm_start
;
1898 if ((PAGE_SIZE
-1) & (unsigned long)addr
)
1901 area
= find_vm_area(addr
);
1905 if (!(area
->flags
& VM_USERMAP
))
1908 if (usize
+ (pgoff
<< PAGE_SHIFT
) > area
->size
- PAGE_SIZE
)
1911 addr
+= pgoff
<< PAGE_SHIFT
;
1913 struct page
*page
= vmalloc_to_page(addr
);
1916 ret
= vm_insert_page(vma
, uaddr
, page
);
1923 } while (usize
> 0);
1925 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1926 vma
->vm_flags
|= VM_RESERVED
;
1930 EXPORT_SYMBOL(remap_vmalloc_range
);
1933 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1936 void __attribute__((weak
)) vmalloc_sync_all(void)
1941 static int f(pte_t
*pte
, pgtable_t table
, unsigned long addr
, void *data
)
1943 /* apply_to_page_range() does all the hard work. */
1948 * alloc_vm_area - allocate a range of kernel address space
1949 * @size: size of the area
1951 * Returns: NULL on failure, vm_struct on success
1953 * This function reserves a range of kernel address space, and
1954 * allocates pagetables to map that range. No actual mappings
1955 * are created. If the kernel address space is not shared
1956 * between processes, it syncs the pagetable across all
1959 struct vm_struct
*alloc_vm_area(size_t size
)
1961 struct vm_struct
*area
;
1963 area
= get_vm_area_caller(size
, VM_IOREMAP
,
1964 __builtin_return_address(0));
1969 * This ensures that page tables are constructed for this region
1970 * of kernel virtual address space and mapped into init_mm.
1972 if (apply_to_page_range(&init_mm
, (unsigned long)area
->addr
,
1973 area
->size
, f
, NULL
)) {
1978 /* Make sure the pagetables are constructed in process kernel
1984 EXPORT_SYMBOL_GPL(alloc_vm_area
);
1986 void free_vm_area(struct vm_struct
*area
)
1988 struct vm_struct
*ret
;
1989 ret
= remove_vm_area(area
->addr
);
1990 BUG_ON(ret
!= area
);
1993 EXPORT_SYMBOL_GPL(free_vm_area
);
1995 static struct vmap_area
*node_to_va(struct rb_node
*n
)
1997 return n
? rb_entry(n
, struct vmap_area
, rb_node
) : NULL
;
2001 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2002 * @end: target address
2003 * @pnext: out arg for the next vmap_area
2004 * @pprev: out arg for the previous vmap_area
2006 * Returns: %true if either or both of next and prev are found,
2007 * %false if no vmap_area exists
2009 * Find vmap_areas end addresses of which enclose @end. ie. if not
2010 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2012 static bool pvm_find_next_prev(unsigned long end
,
2013 struct vmap_area
**pnext
,
2014 struct vmap_area
**pprev
)
2016 struct rb_node
*n
= vmap_area_root
.rb_node
;
2017 struct vmap_area
*va
= NULL
;
2020 va
= rb_entry(n
, struct vmap_area
, rb_node
);
2021 if (end
< va
->va_end
)
2023 else if (end
> va
->va_end
)
2032 if (va
->va_end
> end
) {
2034 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2037 *pnext
= node_to_va(rb_next(&(*pprev
)->rb_node
));
2043 * pvm_determine_end - find the highest aligned address between two vmap_areas
2044 * @pnext: in/out arg for the next vmap_area
2045 * @pprev: in/out arg for the previous vmap_area
2048 * Returns: determined end address
2050 * Find the highest aligned address between *@pnext and *@pprev below
2051 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2052 * down address is between the end addresses of the two vmap_areas.
2054 * Please note that the address returned by this function may fall
2055 * inside *@pnext vmap_area. The caller is responsible for checking
2058 static unsigned long pvm_determine_end(struct vmap_area
**pnext
,
2059 struct vmap_area
**pprev
,
2060 unsigned long align
)
2062 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2066 addr
= min((*pnext
)->va_start
& ~(align
- 1), vmalloc_end
);
2070 while (*pprev
&& (*pprev
)->va_end
> addr
) {
2072 *pprev
= node_to_va(rb_prev(&(*pnext
)->rb_node
));
2079 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2080 * @offsets: array containing offset of each area
2081 * @sizes: array containing size of each area
2082 * @nr_vms: the number of areas to allocate
2083 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2084 * @gfp_mask: allocation mask
2086 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2087 * vm_structs on success, %NULL on failure
2089 * Percpu allocator wants to use congruent vm areas so that it can
2090 * maintain the offsets among percpu areas. This function allocates
2091 * congruent vmalloc areas for it. These areas tend to be scattered
2092 * pretty far, distance between two areas easily going up to
2093 * gigabytes. To avoid interacting with regular vmallocs, these areas
2094 * are allocated from top.
2096 * Despite its complicated look, this allocator is rather simple. It
2097 * does everything top-down and scans areas from the end looking for
2098 * matching slot. While scanning, if any of the areas overlaps with
2099 * existing vmap_area, the base address is pulled down to fit the
2100 * area. Scanning is repeated till all the areas fit and then all
2101 * necessary data structres are inserted and the result is returned.
2103 struct vm_struct
**pcpu_get_vm_areas(const unsigned long *offsets
,
2104 const size_t *sizes
, int nr_vms
,
2105 size_t align
, gfp_t gfp_mask
)
2107 const unsigned long vmalloc_start
= ALIGN(VMALLOC_START
, align
);
2108 const unsigned long vmalloc_end
= VMALLOC_END
& ~(align
- 1);
2109 struct vmap_area
**vas
, *prev
, *next
;
2110 struct vm_struct
**vms
;
2111 int area
, area2
, last_area
, term_area
;
2112 unsigned long base
, start
, end
, last_end
;
2113 bool purged
= false;
2115 gfp_mask
&= GFP_RECLAIM_MASK
;
2117 /* verify parameters and allocate data structures */
2118 BUG_ON(align
& ~PAGE_MASK
|| !is_power_of_2(align
));
2119 for (last_area
= 0, area
= 0; area
< nr_vms
; area
++) {
2120 start
= offsets
[area
];
2121 end
= start
+ sizes
[area
];
2123 /* is everything aligned properly? */
2124 BUG_ON(!IS_ALIGNED(offsets
[area
], align
));
2125 BUG_ON(!IS_ALIGNED(sizes
[area
], align
));
2127 /* detect the area with the highest address */
2128 if (start
> offsets
[last_area
])
2131 for (area2
= 0; area2
< nr_vms
; area2
++) {
2132 unsigned long start2
= offsets
[area2
];
2133 unsigned long end2
= start2
+ sizes
[area2
];
2138 BUG_ON(start2
>= start
&& start2
< end
);
2139 BUG_ON(end2
<= end
&& end2
> start
);
2142 last_end
= offsets
[last_area
] + sizes
[last_area
];
2144 if (vmalloc_end
- vmalloc_start
< last_end
) {
2149 vms
= kzalloc(sizeof(vms
[0]) * nr_vms
, gfp_mask
);
2150 vas
= kzalloc(sizeof(vas
[0]) * nr_vms
, gfp_mask
);
2154 for (area
= 0; area
< nr_vms
; area
++) {
2155 vas
[area
] = kzalloc(sizeof(struct vmap_area
), gfp_mask
);
2156 vms
[area
] = kzalloc(sizeof(struct vm_struct
), gfp_mask
);
2157 if (!vas
[area
] || !vms
[area
])
2161 spin_lock(&vmap_area_lock
);
2163 /* start scanning - we scan from the top, begin with the last area */
2164 area
= term_area
= last_area
;
2165 start
= offsets
[area
];
2166 end
= start
+ sizes
[area
];
2168 if (!pvm_find_next_prev(vmap_area_pcpu_hole
, &next
, &prev
)) {
2169 base
= vmalloc_end
- last_end
;
2172 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2175 BUG_ON(next
&& next
->va_end
<= base
+ end
);
2176 BUG_ON(prev
&& prev
->va_end
> base
+ end
);
2179 * base might have underflowed, add last_end before
2182 if (base
+ last_end
< vmalloc_start
+ last_end
) {
2183 spin_unlock(&vmap_area_lock
);
2185 purge_vmap_area_lazy();
2193 * If next overlaps, move base downwards so that it's
2194 * right below next and then recheck.
2196 if (next
&& next
->va_start
< base
+ end
) {
2197 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2203 * If prev overlaps, shift down next and prev and move
2204 * base so that it's right below new next and then
2207 if (prev
&& prev
->va_end
> base
+ start
) {
2209 prev
= node_to_va(rb_prev(&next
->rb_node
));
2210 base
= pvm_determine_end(&next
, &prev
, align
) - end
;
2216 * This area fits, move on to the previous one. If
2217 * the previous one is the terminal one, we're done.
2219 area
= (area
+ nr_vms
- 1) % nr_vms
;
2220 if (area
== term_area
)
2222 start
= offsets
[area
];
2223 end
= start
+ sizes
[area
];
2224 pvm_find_next_prev(base
+ end
, &next
, &prev
);
2227 /* we've found a fitting base, insert all va's */
2228 for (area
= 0; area
< nr_vms
; area
++) {
2229 struct vmap_area
*va
= vas
[area
];
2231 va
->va_start
= base
+ offsets
[area
];
2232 va
->va_end
= va
->va_start
+ sizes
[area
];
2233 __insert_vmap_area(va
);
2236 vmap_area_pcpu_hole
= base
+ offsets
[last_area
];
2238 spin_unlock(&vmap_area_lock
);
2240 /* insert all vm's */
2241 for (area
= 0; area
< nr_vms
; area
++)
2242 insert_vmalloc_vm(vms
[area
], vas
[area
], VM_ALLOC
,
2249 for (area
= 0; area
< nr_vms
; area
++) {
2261 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2262 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2263 * @nr_vms: the number of allocated areas
2265 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2267 void pcpu_free_vm_areas(struct vm_struct
**vms
, int nr_vms
)
2271 for (i
= 0; i
< nr_vms
; i
++)
2272 free_vm_area(vms
[i
]);
2276 #ifdef CONFIG_PROC_FS
2277 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2280 struct vm_struct
*v
;
2282 read_lock(&vmlist_lock
);
2284 while (n
> 0 && v
) {
2295 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
2297 struct vm_struct
*v
= p
;
2303 static void s_stop(struct seq_file
*m
, void *p
)
2305 read_unlock(&vmlist_lock
);
2308 static void show_numa_info(struct seq_file
*m
, struct vm_struct
*v
)
2311 unsigned int nr
, *counters
= m
->private;
2316 memset(counters
, 0, nr_node_ids
* sizeof(unsigned int));
2318 for (nr
= 0; nr
< v
->nr_pages
; nr
++)
2319 counters
[page_to_nid(v
->pages
[nr
])]++;
2321 for_each_node_state(nr
, N_HIGH_MEMORY
)
2323 seq_printf(m
, " N%u=%u", nr
, counters
[nr
]);
2327 static int s_show(struct seq_file
*m
, void *p
)
2329 struct vm_struct
*v
= p
;
2331 seq_printf(m
, "0x%p-0x%p %7ld",
2332 v
->addr
, v
->addr
+ v
->size
, v
->size
);
2335 char buff
[KSYM_SYMBOL_LEN
];
2338 sprint_symbol(buff
, (unsigned long)v
->caller
);
2343 seq_printf(m
, " pages=%d", v
->nr_pages
);
2346 seq_printf(m
, " phys=%lx", v
->phys_addr
);
2348 if (v
->flags
& VM_IOREMAP
)
2349 seq_printf(m
, " ioremap");
2351 if (v
->flags
& VM_ALLOC
)
2352 seq_printf(m
, " vmalloc");
2354 if (v
->flags
& VM_MAP
)
2355 seq_printf(m
, " vmap");
2357 if (v
->flags
& VM_USERMAP
)
2358 seq_printf(m
, " user");
2360 if (v
->flags
& VM_VPAGES
)
2361 seq_printf(m
, " vpages");
2363 show_numa_info(m
, v
);
2368 static const struct seq_operations vmalloc_op
= {
2375 static int vmalloc_open(struct inode
*inode
, struct file
*file
)
2377 unsigned int *ptr
= NULL
;
2381 ptr
= kmalloc(nr_node_ids
* sizeof(unsigned int), GFP_KERNEL
);
2382 ret
= seq_open(file
, &vmalloc_op
);
2384 struct seq_file
*m
= file
->private_data
;
2391 static const struct file_operations proc_vmalloc_operations
= {
2392 .open
= vmalloc_open
,
2394 .llseek
= seq_lseek
,
2395 .release
= seq_release_private
,
2398 static int __init
proc_vmalloc_init(void)
2400 proc_create("vmallocinfo", S_IRUSR
, NULL
, &proc_vmalloc_operations
);
2403 module_init(proc_vmalloc_init
);